The Slave Trade and Ethnic Stratification in Africa
Warren Whatley and Rob Gillezeau
July 5, 2010
Contact Information
Warren Whatley, Department of Economics, University of Michigan, 611 Tappan Street, Ann Arbor, MI, 48104
([email protected])
Rob Gillezeau, Department of Economics, University of Michigan, 611 Tappan Street, Ann Arbor, MI, 48104
([email protected])
1
Introduction
In a recent article published in the Journal of African History, A. G. Hopkins (2009), author of perhaps
the most-influential book on African economic history, An Economic History of West Africa (1973), argues
that now is the time for a revival of African economic history. “Unknown to most historians,” he argues,
“economists have produced a new economic history of Africa in the course of the past decade,” the two most
important contributions being “the thesis that Africa has suffered a ’reversal of fortune’ during the past 500
years, and the proposition that ethnic fragmentation, which has deep historical roots, is a distinct cause of
African economic backwardness (p. 155).”
In this paper we propose to connect these themes by analyzing the relationship between the slave trade
in the past and ethnic fragmentation in Africa today. Nunn (2008) finds that the transatlantic slave trade
resulted in the long-term, systematic underdevelopment of many African economies. However, this work does
not capture the mechanism through which this underdevelopment may have occurred. Nunn and Wantchekon
(2009) make an effort to explain the process through the development of mistrust as a result of the slave
1
trade. This is a plausible mechanism and we believe that mistrust likely played a significant role in Africa’s
long run economic growth, but we do not believe that this was the only major factor. In a related paper,
Whatley and Gillezeau (2011) show that under plausible conditions the slave trade may have constrained
the geographic scope of authority and increased the salience of ethnic identity. If the slave trade did, in fact,
increase ethnic diversity in Africa then this may have discouraged economic growth as argued by Levine and
Easterly (1997), Collier (1998), Bates (2008) and others. Our goal in this paper is to test whether there
exists empirical evidence that the slave trade increased ethnic diversity.
2
Ethnic Identity and the Slave Trade
A number of important studies have focused on ethnic stratification and its impact on economic performance
in Africa. The best known is a study by Levine and Easterly (1997) which argues that roughly 25% of the
difference in the growth experiences of African and Asian economies can be attributed to the greater ethnic
diversity in Africa. While it is unclear precisely how ethnicity influences economic performance, the authors
present some evidence on a negative relationship between ethnic diversity and under-investment in schooling,
weak financial institutions, poor infrastructure and black-market premia. In a related piece Alesina, Baqir
and Easterly (1999) present evidence that the diversity found in United States cities reduces spending on
public services and increases rent-seeking activities.
Collier (1998) cautions that the relationship between ethnicity and economic performance is more complex
and contextual than this. While arguing and presenting evidence that ethnic diversity can become a drag
on growth, Collier adds the proviso that the negative effects are largely confined to economies with limited
individual rights. In fact, ethnic diversity can be a plus. While democratic institutions can effectively
mitigate the negative effects of ethnic diversity, highly diverse countries are less likely (not more likely) to
break out into ethnic conflict, presumably because of the higher cost of inter-ethnic cooperation. Bates
(2008) contextualizes the impact of ethnicity in a similar way. He argues that the predatory nature of the
postcolonial state in many African countries created political and military challenges to its authority. When
the challenges intensified, ethnic stratification also intensified to the point where “things fell apart.”
2
The literature on ethnic conflict tends to assume that the oppositional character of ethnic identity, with
its insider-outsider distinctions, is the source of conflict that impedes growth. A useful alternative view
is offered by Esteban and Ray (2008). In situations where political behavior can be modeled as “prize
grabbing” mass mobilization, there is a built-in bias towards ethnic rather than class mobilization because
ethnic groups include the rich, who have the resources, and the poor, who provide the labor needed to mount
a mass movement. Conflict will tend to occur along ethnic lines, not because ethnic identity is inherently
conflictual but because it is easier to mount an ethnically-based mass movement.
In all of these examples ethnicity is treated as exogenous and given to the situation. In fact, Collier
expresses an uneasiness about the negative connotations being attached to ethnic diversity in Africa because
“... there is nothing a country can legitimately do about its ethnic composition” (1998, page 387-88). But
there is a large and growing literature which attempts to endogenize ethnic identity, to varying degrees.
This literature tends to emphasize the fact that people have multiple identities that are malleable, politically
manipulable and situational. Posner (2006), for example, develops and tests a model explicitly designed to
identify the conditions under which individual Zambians choose to organize around one particular identity
rather than another. Individuals are viewed as having a portfolio of identities from which they can choose,
and it is postulated that individuals choose the one that has the best chances of putting them in the winning
coalition. The important political choice in post-colonial Zambia is between ethnic identity and language
identity, and Posner is successful in revealing the conditions under which people choose one or the other.
Still, in this formulation ethnic identity as distinct from language identity retains a high degree of exogeniety.
The choice is between ethnic and language identity, not between competing ethnic identities.
Ethnic identity becomes more endogenous and malleable when one leaves the realm of rational choice
and takes a historical view. Posner (2006, pages 21-88) spends two chapters tracing the historical origins of
Zambia ethnic and language groupings. The conventional wisdom here emphasizes the role played by the
institutions of colonial rule, not the conflict and violence of the slave trade. Quoting Posner,
“In tracing the origins of contemporary Zambian ethnic identity to the institutions of colonial
rule, I am following an extremely well-trodden path. In fact, the notion that the colonial state
3
created or heightened the importance of ethnic identities in postcolonial Africa is so accepted
these days that to argue otherwise would probably be controversial (2006, page 23).”
1
Yet otherwise is precisely what we want to argue. The conventional view roots the salience of ethnic identity in Africa in what Firmin-Sellers calls “the logic of indirect rule” (1996, 2000). Colonial administrations,
befuddled by the variety of local ethnic political economies they encountered, found it difficult to extract
economic surplus directly. In situations like this, characterized by asymmetric information, the principal (the
colonial power) has an incentive to share the surplus with agents (indigenous authorities) who know how to
monitor and direct the production and flow of surplus to the top. In areas where the indigenous political
authorities were large and strong, the colonial powers enlisted them. Where the indigenous authorities were
small, weak or non-existent, the colonial powers created them. In either case, the colonial power stood
behind and strengthened the indigenous territorial authorities, often drawing maps to clearly delineate the
boundaries of these ethnicities. Posner (2006) argues that the logic of indirect rule also provided incentives
for local inhabitants to identify with the prevailing social prescriptions that legitimize the local authority.
It is through this identity – this ethnic identity – that local inhabitants gained access to important public
goods.
This view is plausible and well-documented. The point we want to make in this paper is that the slave
trade helped shape the ethnic landscape that the colonial powers encountered in Africa. We are not trying
to overturn the conventional wisdom but to root it more firmly in the history of Africa. In fact, we use the
many maps of ethnic boundaries drawn by colonial authorities to construct our measure of ethnic diversity
across the African landscape. We then ask did the intensity of past slaving activities influence the ethnic
landscape that emerged out of the colonial era? Our prior, formulate in Whatley and Gillezeau (2011), is
that the slave trade influenced the spatial distribution of political authority and the salience of ethnic identity
that so befuddled the colonial authorities and forced them into indirect rule. In that paper we argue that
when the international demand for Africans as slaves penetrates an area, the value of people to slave raiders
exceeds their value to states as citizens to be protected and taxed. State expansion slows. Slave raiding
1 In
addition to the long list of studies cited in Posner (2006, page 23, footnote 1) one could add Peel (1989), O’Brien (1986),
Ranger (1996) and Kaarsholm and Hultin (1994).
4
intensifies. The cost of protecting citizens increases. The benefits of distinguishing insider from outsider
increases. There is an increased incentive to reproduce “others” who can be raided. All of these forces
contribute to a greater degree of ethnic diversity across the African landscape. We believe that recognition
of a history of slaving in Africa can help explain the salience of ethnic identity among African people, the
great diversity of ethnic identities on the continent of Africa and the geographic distribution and sizes of
these ethnicities.
3
Emprical Strategy
In order to determine the impact of the transatlantic slave trade on the long-run development of ethnicity
in Africa, we need to compare the number of ethnicities in equally sized regions along the West African
coast and the number of slaves that departed from these same regions. Our basic strategy is as follows. We
divide the western coast of Africa into evenly spaced points numbering 200 in total.2 The points start at the
northernmost point of Tunisia and end at the middle of South Africa. The distance between these points
is 50 kilometres.3 These coastal points serve as the basis for our analysis
4
and both the dependent and
independent variables are constructed using buffer zones around them. Our dependent variable is the number
of ethnicities (or ethno-linguistic subgroups) in the region around each coastal point. This ethno-linguistic
data is taken from Felix and Meur (2001), having been digitized by the Harvard AfricaMap group (2010). It
is our understanding that this is the most modern Africa-wide ethno-linguistic classification map available.
For robustness, we also use the ethno-linguistic mapping of Africa by Murdock (1959) to assign ethnicities
to each coastal point.5 This is not our preferred measure, however, as it stifles much of the variation in
more modern mappings and appears to group sub-ethnicities together. Given the relatively short time span
involved in our analysis, and given the likelihood that it takes generations for ethnicities to be generated
or to disappear we require as fine a measure as possible. Our independent variables include the number
of slaves exported from nearby African ports, courtesy of the Transatlantic Slave Trade Database (2010),
2 We
also perform our analysis with the coast divided into a total of 50 or 100 points.
3 When
working with 100 observations the distance is 100 kilomters and when working with 50 observations it is 200 kilometers.
4 Refer
to Figure 2 in the Appendix for a visual representation.
5 Refer
to Figure 1 in the Appendix for a visual representation.
5
local agricultural suitability (Fischer et al. 2002) as measured by climate, soil and terrain slope constraints,
population density in 1960 (UNESCO 1987), elevation (USGS 2010), forest coverage (Fischer et al. 2002),
and desert coverage (Fischer et al. 2002).
We perform our regression analysis with 3 different circular buffer sizes (and one none-buffer method):
125 kilometres, 250 kilometres, and 500 kilometres. This means that in our analysis using the 125 kilometre
buffer our environmental variables are based on their mean value in that region, the number of ethnicities is
the total found within that buffer, and slave exports are the total exported from slaving ports within that
buffer. In addition to the buffer method, as a robustness check, we also perform our analysis assigning each
ethnicity to the nearest of our equally spaced coastal points (we apply the 500 kilometre buffer when using
this nearest technique). Using each of these methods, we perform the following OLS regression:
Ei = α + β1 Si + γXi + vi
Where Ei is the number of ethnicities assigned to coastal point i, α is the intercept, Si is the number of
slave exports assigned to coastal point i, Xi is a vector of environmental covariates assigned to coastal point
i, and vi is an error term.
As the reader has likely noted, there is almost certainly some degree of reverse causation in the above
specifications. If slaving was taboo within one’s own ethnic group it would have been necessary for other
ethnicities to be present in order to capture slaves. Indeed, the greater the number of regional ethnic groups,
the larger the potential for slaving. In order to present a causal estimate of the impact of slaving on the
development of ethnicity, we make use of the instruments developed by Nunn (2008) which are composed of
the distance of region to the regional slaving markets. We do not, however, make use of all four of Nunn’s
instruments at the present. Rather, we include the instruments based on the minimum distance to the
Atlantic trade (Virginia, Havana, Haiti, Kingston, Dominica, Guyana, Salvador, Rio de Janeiro) and the
Trans-Saharan trade (Algiers, Tunis, Tripoli, Bengahzi, Cairo). Given that Nunn’s instrument has been
accepted as valid in his work, combined with the understanding in advance of his work that ethnicity may
discourage economic growth [Levine and Easterly, 1997], we may conclude that the instrument remains valid
6
for our purposes.
In future work, we intend to extend our analysis to regions further inland under the assumption that
fewer slaves are acquired in the interior. We also intend to shift from circular buffers to rectangular buffers
without the possbility for overlap.
4
Results
In the summary table below we present results from the array of regression specifications described in the
prior section. The full set of regressions using the Peoples Atlas for the measure of ethnicity is in Tables 2-5
of the Appendix. The complete set of regressions based on Murdock’s measure of ethnicity is in Tables 6-9.
The first stage IV results for all regressions in Tables 1-9 are in tables 10-12.
Type
OLS NC
OLS NC
OLS NC
OLS
OLS
OLS
IV
IV
IV
Peoples - 125km
.021∗∗∗
.020∗∗∗
.022∗∗∗
.015∗∗∗
.013∗∗∗
.014∗∗
.038∗∗∗
.038∗∗
.041
Peoples - 250km
.072∗∗∗
.072∗∗∗
.078∗∗∗
.041∗∗∗
.042∗∗∗
.028∗
.123∗∗∗
.120∗∗∗
.141∗∗
Peoples - 500km
.169∗∗∗
.173∗∗∗
.179∗∗∗
.160∗∗∗
.166∗∗∗
.170∗
.342∗∗∗
.353∗∗∗
.407∗∗
∗∗∗
∗∗∗
∗∗∗
∗∗∗
∗∗∗
Peoples - Nearest
Murdock - 125km
Murdock - 250km
Murdock - 500km
.028
∗
.001
∗∗∗
.009
∗∗∗
.021
∗∗∗
.019
.001
∗∗∗
.009
∗∗∗
.022
∗∗∗
.024
.002
∗∗∗
.009
∗∗∗
.023
∗∗∗
.025
−.0004
∗∗
.003
∗∗∗
.019
∗∗∗
.018
.009
∗∗
.059
∗∗∗
∗∗∗
.068
∗∗
.047
−.0007
−.0002
.038
.038
.041
.002
−.0008
.0006
.002
−.003
∗∗∗
.021
∗∗∗
∗∗∗
.019
∗
∗∗∗
.021
∗∗∗
.0002
∗∗∗
.024
Murdock - Nearest
.004
.002
.003
.004
.003
.002
.007
.006
.010
Obs
200
100
50
200
100
50
200
100
50
Table 1: This table presents the coefficient on slave exports over 72 specifications. The results presented in
this table are calculated using OLS or 2SLS, as marked. OLS NC indicates that there are no environmental
controls. The units for slave exports are thousands of people. The variables are constructed in a 125km,
250km, or 500km buffer around each of the coastal points as marked. The “nearest” specification uses a
500km buffer for environmental characteristics. Specifications include totals of: 50, 100, and 200 total points.
The measure of ethnicity is constructed using Felix and Meur (2001) or Murdock (1959). Refer to Appendix
Tables 2-12 for complete results
From the first half of this summary table, it is clear that there is a robust, positive relationship between
7
regional slave exports and number of ethnicities as defined by the Peoples Atlas. This result holds in the
basic specification, with the full set of environmental controls, and using the instrumental variables strategy.
Amongst the controls agricultural suitability and total population are consistently positively related to the
number of ethnicities while elevation, forest, and desert cover are all negatively related to the number of
ethnicities. In the second half of the table, we see that the positive relationship between slave exports
and ethnicity tends to persist using the measure generated by Murdock (1959), although it is a weaker
relationship than with the Peoples Atlas.6 In general, the results are stronger the greater buffer zone and
the greater the number of observations (meaning that there is a smaller distance between observations). The
results are robust for all buffer sizes and numbers of observations for the Peoples Atlas and the Murdock
results are generally robust, although weaker for smaller buffer zones. The results are robust to removing
most observations from North Africa and South Africa, however, they lose significance as the span of coast
is shrunk.
As to the size of the coefficients, the small buffer zone estimates using ethnicities from the Peoples Atlas
indicate that the slave trade resulted in an average increase of 0.9 to 2.3 local ethnicities in each of the 200
coastal regions. The larger buffer zone estimates, using the Peoples atlas, suggest an average local increase
(over a much larger area) of 43.6 to 110.95 ethnicities. Since, ethnicities overlap across coastal points this
overstates the treatment effect. However, the Peoples Atlas contains roughly 3700 ethnicities so this is still
an economically significant effect.
5
Discussion and Conclusion
In this paper we have argued that the slave trade constrained the geographic scope of political authority
and heightened the incentive to distinguish insider from outsider. OLS regressions identify a positive and
statistically significant relationship between the number of slaves leaving the west coast of Africa in the
past and the limited geographic scope of twentieth century ethnic groupings. This relationship is robust to
changes in the scheme for drawing ethnic boundaries and to the inclusion of a variety of variables thought to
6 Murdock’s
mapping only includes a fraction of the ethnicities present Felix and Meur’s map
8
influence the geographic scope of ethnic groupings. Instrumental variable estimations produce support for
the view that causality runs from slaving to ethnic diversity.
We believe this finding has broad implications for research in the economic history of Africa.
Nunn and Wantchekon (2009) find evidence that the intensity of slave capturing and marketing in the
past helps explain spatial and individual variations in the level of mistrust among Africans today. Coupled
with the evidence on ethnic conflict and its salience in Africa, one might expect mistrust to be one of the
many social manifestations of the kind of heightened ethnic identity that we find correlated with the slave
trade.
Lonsdale (1994) has emphasized the difference between what he calls “moral ethnicity” and “political
tribalism.” Political tribalism is the rational prize-grabbing that characterizes much of the literature in
political science. Moral ethnicity is that contested internal standard of civic virtue against which is measured
personal esteem. We believe it is this latter set of social relations recorded by ethnographers that are mostinfluenced by the slave trade (see, for example, Soumonnu (2003)and Murdock (1959, page 253)). Distinctions
may have been exaggerated by anthropologists and intensified by colonial administrations, but we present
evidence that the spatial variations in the sizes of these authority systems is negatively related to the intensity
of slaving in the past.
Ackerlof and Kranton (2000, 2010) have modeled something similar to moral ethnicity. They call it an
economics of identity, where individuality can produce behaviors that run counter to social prescriptions,
but which illicit punitive responses from others whose identity has been called into question by the deviant
behavior. The geographic concentration of moral ethnicity and the closed nature of the social systems
described by ethnographers are the ideal settings for this kind of social interaction. Their formulation
identifies several policies that could reduce ethnic conflict, the most important being policies that increase
the cost of penalizing deviant behavior.
At the most general level, our findings endogenize some of the ethnic diversity that characterizes contemporary Africa. Rather than view the salience of ethnic identity in Africa as something primordial, traditional,
or even primitive, this paper presents evidence that it is quite the opposite – a legacy of the role and position
9
of Africa in the creation of our modern world.
At the same time, it is consistent with the view that ethnic diversity has roots in Africa that are deeper
than the colonial experience. In this sense, it helps explain why colonial powers were forced into indirect
rule and the strengthening of “traditional” authority. The plethora of moral ethnicities they encountered
constrained the effectiveness of direct rule and prevented the wholesale importation of European institutions.
Acemoglu et. al. (2000) would see this as a reversal of fortune. In this particular case, the extractive
institution is organized slave raiding, which Nunn (2008) argue is not conducive to long-run growth. What
we add to this line of thinking is a lock in mechanism – ethnic diversity – which locks-out the importation
of an alternative set of institutions that may have been more favorable to growth.
Our next step is to ask how slaving influenced the structure of governance in Africa. Chiefdoms and
kingdoms were the most prevalent forms of governance that extended beyond the village. Vansina (2004)
has written a fascinating book on the emergence of governance in South Central Africa, combining his
remarkable expertise in historical linguistics, archeology, geography and history. While the argument is much
too complex to summarize here, a fruitful line of investigation is the idea that endowments (cattle, gold, soil
quality, etc.), technologies (agriculture, water-wells, etc) and forms of wealth (size and mobility)influence
social relations (like inheritance, marriage and residence rules), which in turn influence the structure and
territoriality of governance. An obvious example is how matrilineal descent with patrilocal residence expands
the geographic scope of chiefly authority through clan connections. At the other extreme is how patrilineal
descent with patrilocal residence strengthens kin-based wealth accumulation but weakens the territorial scope
of governance. It is not clear how the slave trade would influence these complex structures of governance,
but one could pursue this empirically along the line first proposed by Paden (1980) and Kaufert (1980) who
use the Murdock data files to look for “culture cluster” underlying his ethnic groupings. One could just as
easily look for “governance cluster” and see if they are influenced by a history of slaving. We plan to pursue
this line of research.
10
6
Works Cited
Acemoglu, D., S. Johnson, et al. (2000). The colonial origins of comparative development : an
empirical investigation. Cambridge, MA, National Bureau of Economic Research.
Akerlof, G. A. and R. E. Kranton (2000). Economics and Identity. The Quarterly Journal of
Economics CXV(3): 715-753.
Akerlof, G. A. and R. E. Kranton (2010). Identity Economics: How Our Identities Shape Our Work,
Wages and Well-Being. Princeton, Princeton University Press.
Alesina, A., R. Baqir, et al. (1999). Public Goods and Ethnic Divisions. The Quarterly Journal of
Economics 114(4): 1243-1284.
Bates, R. H. (2008). When things fell apart: state failure in late-century Africa. New York Cambridge
University Press.
Collier, P. (1998). The Political Economy of Ethnicity. Annual World Bank Conference on Development
Economics. B. Preskovic and J. E. Stigletz. Washington, D. C., The World Bank: 387-399.
Eltis, D. (2010). The Trans-Atlantic Slave Trade Database..
Esteban, J. and D. Ray (2008). On the Salience of Ethnic Conflict. American Economic Review 95(5):
2185-2202.
Firmin-Sellers, K. (1996). The Transformation of Property Rights in the Gold Coast. Cambridge,
Cambridge University Press.
Fischer, G., Sharh, M. Velthuizen, H., and Nachtergaele, F.O. (2002). Global Agro-Ecological
Assessment for Agriculture in the 21st Century: Methodology and Results. International Institute for
Applied Systems Analysis/Food and Agriculture Organization of the United Nations, Laxenburg and Rome.
Firmin-Sellers, K. (2000). Institutions, Context, and Outcomes: Explaining French and British Rule in
West Africa. Comparative Politics 32(3): 253-272.
Hay, R. Jr. and J. Paden (1980). A Culture Cluster Analysis of Six African States. in Values,
11
Identities and National Integration: empirical research in Africa. John Paden. Evanston, Northwestern
University Press.
Hopkins, A. G. (1973). An economic history of West Africa. New York, Columbia University Press.
Hopkins, A. G. (2009). The New Economic History of Africa. Journal of African History 50(1): 155-177.
Kaarsholm, P. and J. Hultin (1994). Inventions and Boundaries: historical and anthropological
approaches to the study of ethnicity and nationalism. Roskilde University, Institute for Development
Studies.
Kaufert, J. M. (1980). Ethnic Unit Definition in Ghana: a comparison of culture clusters analysis and
social distance measures. in Values, Identities and National Integration: empirical research in Africa. John
Paden. Evanston, Northwestern University Press.
Levine, R. and W. Easterly (1997). Africa’s Growth Tragedy: Policies and Ethnic Divisions The
Quarterly Journal of Economics 112(4): 1203-1250
Lonsdale, J. (1994). Moral Ethnicity and Political Tribalism. Inventions and Boundaries: historical and
anthropological approaches to the study of ethnicity and nationalism. P. Kaarsholm and J. Hultin. Roskilde
University, Institute for Development Studies.
Murdock, G. P. (1959). Africa: Its People and their Culture. New York, McGraw-Hill.
Nunn, N. (2008). The Long Term Effects of Africa’s Slave Trades. Quarterly Journal of Economics
123(1): 139-176.
Nunn, N. and L. Wantchekon (2009). The Slave Trade and the Origins of Mistrust in Africa.
O’Brien, J. (1986). Towards a Reconstruction of Ethnicity: Capitalist Expansion and Cultural Dynamics
in Sudan. American Anthropologist 88: 898-906.
Peel, J. D. Y. (1996). The Cultural Work of Yoruba Ethnogenesis. History and Ethnicity. E. Tonkin, M.
McDonald and M. Chapman. London, Routledge.
Posner, D. N. (2006). Institutions and Ethnic Politics in Africa. New York, Cambridge University Press.
12
Ranger, T. (1996). PostScript: Colonial and Postcolonial Identities. Postcolonial Identities in Africa. T.
Ranger. London, Zed Books.
UNESCO (1987). UNEP/GRID - Sioux Falls Clearninghouse http://na.unep.net/datasets/datalist.php.
USGS (2010). USGS Geographic Data Download http://edc2.usgs.gov/geodata/index.php.
Vansina, J. (2004)How Societies Are Born: governence in west central Africa before 1600. Charlottesville,
University of Virginia Press.
Whatley, W. and R. Gillezeau (Forthcoming 2011). The Fundamental Impact of the Slave Trade on
African Economies. Economic Evolution and Revolution in Historical Time. P. Rhode, J. Rosenbloom and
D. Weiman. Stanford, Stanford University Press.
13
7
Appendix
Table 2 - Peoples Atlas - 125km Buffer
(1)
(2)
(3)
(4)
(5)
(6)
IV
(7)
IV
(8)
IV
(9)
.021
(.004)∗∗∗
.020
(.005)∗∗∗
.022
(.007)∗∗∗
.015
(.003)∗∗∗
.013
(.005)∗∗∗
.014
(.007)∗∗
.038
(.012)∗∗∗
.038
(.017)∗∗
.041
(.026)
AgSuitability
3.647
(.653)∗∗∗
3.541
(.900)∗∗∗
3.335
(1.311)∗∗
3.431
(.737)∗∗∗
3.325
(1.037)∗∗∗
3.681
(1.546)∗∗
Population
.076
(.026)∗∗∗
.066
(.034)∗
.080
(.041)∗
.056
(.031)∗
.046
(.041)
.077
(.048)
Elevation
-.087
(.115)
-.194
(.166)
-.548
(.254)∗∗
.080
(.153)
.004
(.230)
-.346
(.348)
Forest
-7.444
(5.355)
-5.744
(7.178)
-7.499
(10.893)
-7.741
(5.984)
-3.860
(8.283)
-12.427
(13.382)
Desert
-15.247
(2.351)∗∗∗
-15.238
(3.372)∗∗∗
-13.652
(4.932)∗∗∗
-12.314
(3.010)∗∗∗
-11.954
(4.420)∗∗∗
-11.439
(6.053)∗
200
100
50
200
100
50
Slaves
Obs.
200
100
50
Table 2: The results presented in this table are calculated using OLS or 2SLS, as marked. The variables are
constructed in a 125km buffer around each of the coastal points. Specifications include totals of: 50, 100,
and 200 total points. The measure of ethnicity is constructed using Felix and Meur (2001).
Table 3 - Peoples Atlas - 250km Buffer
(1)
(2)
(3)
(4)
(5)
(6)
IV
(7)
IV
(8)
IV
(9)
.072
(.008)∗∗∗
.072
(.011)∗∗∗
.078
(.017)∗∗∗
.041
(.007)∗∗∗
.042
(.010)∗∗∗
.028
(.015)∗
.123
(.028)∗∗∗
.120
(.035)∗∗∗
.141
(.061)∗∗
AgSuitability
10.387
(1.937)∗∗∗
9.939
(2.710)∗∗∗
7.871
(3.993)∗∗
6.436
(2.806)∗∗
6.310
(3.785)∗
.356
(7.062)
Population
.885
(.100)∗∗∗
.909
(.154)∗∗∗
1.209
(.221)∗∗∗
.680
(.145)∗∗∗
.631
(.228)∗∗∗
.838
(.379)∗∗
Elevation
-.800
(.267)∗∗∗
-.799
(.386)∗∗
-1.067
(.525)∗∗
-.165
(.401)
-.073
(.580)
.103
(.979)
Forest
-12.314
(6.793)∗
-6.817
(9.282)
5.105
(14.779)
-25.222
(9.706)∗∗∗
-19.858
(13.071)
-3.351
(22.531)
Desert
-17.295
(3.432)∗∗∗
-15.296
(4.402)∗∗∗
-14.335
(8.285)∗
-5.867
(5.767)
-5.739
(6.905)
4.425
(15.554)
200
100
50
200
100
50
Slaves
Obs.
200
100
50
Table 3: The results presented in this table are calculated using OLS or 2SLS, as marked. The variables are
constructed in a 250km buffer around each of the coastal points. Specifications include totals of: 50, 100,
and 200 total points. The measure of ethnicity is constructed using Felix and Meur (2001).
14
Table 4 - Peoples Atlas - 500km Buffer
(1)
(2)
(3)
(4)
(5)
(6)
IV
(7)
IV
(8)
IV
(9)
.169
(.012)∗∗∗
.173
(.018)∗∗∗
.179
(.026)∗∗∗
.160
(.015)∗∗∗
.166
(.021)∗∗∗
.170
(.036)∗∗∗
.342
(.053)∗∗∗
.353
(.074)∗∗∗
.407
(.207)∗∗
3.205
(5.724)
-2.238
(7.580)
1.948
(14.707)
-20.801
(9.983)∗∗
-20.906
(12.388)∗
-39.345
(40.693)
4.369
(.488)∗∗∗
4.612
(.688)∗∗∗
4.850
(1.265)∗∗∗
3.419
(.699)∗∗∗
3.462
(1.027)∗∗∗
4.602
(1.804)∗∗
-.338
(.836)
-.022
(1.173)
-.170
(1.767)
2.830
(1.401)∗∗
2.947
(1.934)
3.315
(3.869)
Forest
-53.943
(10.754)∗∗∗
-43.352
(14.784)∗∗∗
-61.010
(24.603)∗∗
-119.394
(22.658)∗∗∗
-114.883
(33.058)∗∗∗
-124.344
(63.946)∗
Desert
-10.985
(4.043)∗∗∗
-5.510
(4.542)
-16.737
(13.956)
12.465
(8.281)
12.459
(9.036)
33.437
(46.850)
200
100
50
200
100
50
Slaves
AgSuitability
Population
Elevation
Obs.
200
100
50
Table 4: The results presented in this table are calculated using OLS or 2SLS, as marked. The variables are
constructed in a 500km buffer around each of the coastal points. Specifications include totals of: 50, 100,
and 200 total points. The measure of ethnicity is constructed using Felix and Meur (2001).
Table 5 - Peoples Atlas - Nearest
(1)
(2)
(3)
(4)
(5)
(6)
IV
(7)
IV
(8)
IV
(9)
.028
(.007)∗∗∗
.019
(.003)∗∗∗
.024
(.006)∗∗∗
.025
(.010)∗∗
.018
(.005)∗∗∗
.009
(.008)
.059
(.028)∗∗
.047
(.015)∗∗∗
.068
(.050)
AgSuitability
3.674
(3.889)
.437
(1.733)
6.903
(3.422)∗∗
-.792
(5.233)
-2.534
(2.473)
-3.456
(9.847)
Population
.560
(.331)∗
.537
(.157)∗∗∗
1.026
(.294)∗∗∗
.383
(.366)
.354
(.205)∗
.963
(.436)∗∗
Elevation
.054
(.568)
.126
(.268)
-.204
(.411)
.644
(.734)
.599
(.386)
.671
(.936)
Forest
-8.257
(7.306)
-.749
(3.381)
-7.815
(5.725)
-20.431
(11.877)∗
-12.134
(6.600)∗
-23.704
(15.474)
Desert
-2.688
(2.747)
.416
(1.039)
-7.952
(3.248)∗∗
1.674
(4.341)
3.275
(1.804)∗
4.636
(11.337)
200
100
50
200
100
50
Slaves
Obs.
200
100
50
Table 5: The results presented in this table are calculated using OLS or 2SLS, as marked. The environmental
variables are constructed in a 500km buffer around each of the coastal points. Ethnicities are only assigned
to the nearest coastal point. Specifications include totals of: 50, 100, and 200 total points. The measure of
ethnicity is constructed using Felix and Meur (2001).
15
Table 6 - Murdock - 125km Buffer
(1)
(2)
(3)
(4)
(5)
(6)
IV
(7)
IV
(8)
IV
(9)
.001
(.0007)∗
.001
(.001)
.002
(.001)
-.0004
(.0006)
-.0007
(.001)
-.0002
(.001)
-.001
(.002)
-.001
(.003)
-.0009
(.005)
.452
(.127)∗∗∗
.595
(.189)∗∗∗
.463
(.267)∗
.457
(.128)∗∗∗
.600
(.191)∗∗∗
.454
(.274)∗
Population
.009
(.005)∗
.012
(.007)∗
.015
(.008)∗
.009
(.005)∗
.013
(.008)∗
.016
(.008)∗
Elevation
-.049
(.022)∗∗
-.079
(.035)∗∗
-.077
(.052)
-.053
(.027)∗∗
-.083
(.042)∗
-.083
(.062)
Forest
1.891
(1.041)∗
1.470
(1.503)
.040
(2.218)
1.898
(1.043)∗
1.430
(1.523)
.173
(2.370)
Desert
-2.485
(.457)∗∗∗
-2.883
(.706)∗∗∗
-2.496
(1.004)∗∗
-2.558
(.525)∗∗∗
-2.952
(.813)∗∗∗
-2.556
(1.072)∗∗
200
100
50
200
100
50
Slaves
AgSuitability
Obs.
200
100
50
Table 6: The results presented in this table are calculated using OLS or 2SLS, as marked. The variables are
constructed in a 125km buffer around each of the coastal points. Specifications include totals of: 50, 100,
and 200 total points. The measure of ethnicity is constructed using Murdock (1959.
Table 7 - Murdock - 250km Buffer
(1)
(2)
(3)
(4)
(5)
(6)
IV
(7)
IV
(8)
IV
(9)
.009
(.001)∗∗∗
.009
(.002)∗∗∗
.009
(.003)∗∗∗
.003
(.001)∗∗
.002
(.002)
-.0008
(.002)
.0006
(.003)
.002
(.004)
-.003
(.006)
AgSuitability
1.809
(.286)∗∗∗
1.780
(.417)∗∗∗
2.132
(.592)∗∗∗
1.904
(.325)∗∗∗
1.822
(.456)∗∗∗
2.258
(.705)∗∗∗
Population
.127
(.015)∗∗∗
.128
(.024)∗∗∗
.147
(.033)∗∗∗
.132
(.017)∗∗∗
.132
(.028)∗∗∗
.154
(.038)∗∗∗
Elevation
-.182
(.039)∗∗∗
-.192
(.060)∗∗∗
-.253
(.078)∗∗∗
-.197
(.046)∗∗∗
-.201
(.070)∗∗∗
-.272
(.098)∗∗∗
Forest
-.685
(1.004)
.173
(1.430)
1.182
(2.193)
-.374
(1.122)
.324
(1.576)
1.324
(2.250)
Desert
-3.363
(.507)∗∗∗
-3.107
(.678)∗∗∗
-4.068
(1.229)∗∗∗
-3.639
(.667)∗∗∗
-3.218
(.832)∗∗∗
-4.384
(1.553)∗∗∗
200
100
50
200
100
50
Slaves
Obs.
200
100
50
Table 7: The results presented in this table are calculated using OLS or 2SLS, as marked. The variables are
constructed in a 250km buffer around each of the coastal points. Specifications include totals of: 50, 100,
and 200 total points. The measure of ethnicity is constructed using Murdock (1959.
16
Table 8 - Murdock - 500km Buffer
(1)
(2)
(3)
(4)
(5)
(6)
IV
(7)
IV
(8)
IV
(9)
.021
(.002)∗∗∗
.022
(.003)∗∗∗
.023
(.004)∗∗∗
.019
(.002)∗∗∗
.021
(.003)∗∗∗
.019
(.005)∗∗∗
.021
(.006)∗∗∗
.024
(.009)∗∗∗
.0002
(.025)
AgSuitability
2.230
(.898)∗∗
1.164
(1.209)
3.412
(2.242)
2.011
(1.176)∗
.833
(1.457)
6.679
(4.938)
Population
.748
(.077)∗∗∗
.768
(.110)∗∗∗
.782
(.193)∗∗∗
.739
(.082)∗∗∗
.747
(.121)∗∗∗
.801
(.219)∗∗∗
Elevation
-.473
(.131)∗∗∗
-.373
(.187)∗∗
-.479
(.269)∗
-.444
(.165)∗∗∗
-.321
(.227)
-.755
(.469)
Forest
-10.703
(1.687)∗∗∗
-8.867
(2.357)∗∗∗
-12.918
(3.750)∗∗∗
-11.301
(2.669)∗∗∗
-10.133
(3.889)∗∗∗
-7.907
(7.759)
Desert
-2.555
(.634)∗∗∗
-1.460
(.724)∗∗
-4.845
(2.127)∗∗
-2.340
(.976)∗∗
-1.142
(1.063)
-8.815
(5.685)
200
100
50
200
100
50
Slaves
Obs.
200
100
50
Table 8: The results presented in this table are calculated using OLS or 2SLS, as marked. The variables are
constructed in a 500km buffer around each of the coastal points. Specifications include totals of: 50, 100,
and 200 total points. The measure of ethnicity is constructed using Murdock (1959.
Table 9 - Murdock - Nearest
(1)
(2)
(3)
(4)
(5)
(6)
IV
(7)
IV
(8)
IV
(9)
.004
(.001)∗∗∗
.002
(.0006)∗∗∗
.003
(.0009)∗∗∗
.004
(.002)∗∗
.003
(.0008)∗∗∗
.002
(.001)∗
.007
(.005)
.006
(.002)∗∗∗
.010
(.007)
AgSuitability
.597
(.666)
-.176
(.292)
.801
(.484)∗
.156
(.879)
-.491
(.377)
-.621
(1.370)
Population
.101
(.057)∗
.076
(.026)∗∗∗
.150
(.042)∗∗∗
.083
(.062)
.057
(.031)∗
.142
(.061)∗∗
Elevation
-.019
(.097)
.017
(.045)
-.042
(.058)
.039
(.123)
.067
(.059)
.079
(.130)
Forest
-1.803
(1.251)
-.142
(.569)
-1.939
(.809)∗∗
-3.006
(1.996)
-1.346
(1.007)
-4.121
(2.154)∗
Desert
-.272
(.470)
.435
(.175)∗∗
-.747
(.459)
.159
(.729)
.737
(.275)∗∗∗
.982
(1.578)
200
100
50
200
100
50
Slaves
Obs.
200
100
50
Table 9: The results presented in this table are calculated using OLS or 2SLS, as marked. The environmental
variables are constructed in a 500km buffer around each of the coastal points. Ethnicities are only assigned
to the nearest coastal point. Specifications include totals of: 50, 100, and 200 total points. The measure of
ethnicity is constructed using Murdock (1959).
17
Table 10 - IV First Stage - 125km Buffer
(1)
-2.788
(1.725)
(2)
-2.663
(2.520)
(3)
-1.829
(3.442)
Instrument2
-3.059
(.687)∗∗∗
-3.071
(.992)∗∗∗
-2.709
(1.303)∗∗
AgSuitability
-6.889
(15.199)
-5.125
(21.453)
-26.137
(30.270)
Population
2.010
(.590)∗∗∗
1.967
(.819)∗∗
.845
(.933)
Elevation
-7.645
(2.803)∗∗∗
-8.758
(4.223)∗∗
-8.334
(6.499)
Forest
75.342
(112.346)
-32.831
(154.868)
200.407
(227.701)
Desert
-50.328
(50.972)
-59.929
(74.878)
-23.442
(106.040)
200
100
50
Instrument1
Obs.
Table 10: The results presented in this table are the first stage results for all 125km buffer IV regressions.
The environmental variables are constructed in a 125km buffer around each of the coastal points. Ethnicities
are only assigned to the nearest coastal point. Specifications include totals of: 50, 100, and 200 total points.
Table 11 - IV First Stage - 250km Buffer
(1)
-2.788
(1.725)
(2)
-2.663
(2.520)
(3)
-1.829
(3.442)
Instrument2
-3.059
(.687)∗∗∗
-3.071
(.992)∗∗∗
-2.709
(1.303)∗∗
AgSuitability
-6.889
(15.199)
-5.125
(21.453)
-26.137
(30.270)
Population
2.010
(.590)∗∗∗
1.967
(.819)∗∗
.845
(.933)
Elevation
-7.645
(2.803)∗∗∗
-8.758
(4.223)∗∗
-8.334
(6.499)
Forest
75.342
(112.346)
-32.831
(154.868)
200.407
(227.701)
Desert
-50.328
(50.972)
-59.929
(74.878)
-23.442
(106.040)
200
100
50
Instrument1
Obs.
Table 11: The results presented in this table are the first stage results for all 250km buffer IV regressions.
The environmental variables are constructed in a 250km buffer around each of the coastal points. Ethnicities
are only assigned to the nearest coastal point. Specifications include totals of: 50, 100, and 200 total points.
18
Table 12 - IV First Stage - 500km Buffer
(1)
-2.788
(1.725)
(2)
-2.663
(2.520)
(3)
-1.829
(3.442)
Instrument2
-3.059
(.687)∗∗∗
-3.071
(.992)∗∗∗
-2.709
(1.303)∗∗
AgSuitability
-6.889
(15.199)
-5.125
(21.453)
-26.137
(30.270)
Population
2.010
(.590)∗∗∗
1.967
(.819)∗∗
.845
(.933)
Elevation
-7.645
(2.803)∗∗∗
-8.758
(4.223)∗∗
-8.334
(6.499)
Forest
75.342
(112.346)
-32.831
(154.868)
200.407
(227.701)
Desert
-50.328
(50.972)
-59.929
(74.878)
-23.442
(106.040)
200
100
50
Instrument1
Obs.
Table 12: The results presented in this table are the first stage results for all 500km buffer and “nearest”
IV regressions. The environmental variables are constructed in a 125km buffer around each of the coastal
points. Ethnicities are only assigned to the nearest coastal point. Specifications include totals of: 50, 100,
and 200 total points.
19
Figure 1 - Slave Ports and Murdock Ethnicities
Figure 1: This figure displays a snapshot of the ArcGIS map used to create the dataset. Red stars signal
the presence of an historical slave port. Grey lines mark divisions between ethnicities (based on Murdock
(1959) and black lines mark divisions between nation states.
20
Figure 2 - Countries and Coastal Points
Figure 2: This figure also displays the ArcGIS map used to create the dataset. The red circles designate the
coastal points used as our observations in the empirical analysis.
21